Diffusion-Based Bayesian Cluster Enumeration in Distributed Sensor Networks
Freweyni K. Teklehaymanot, Michael Muma, Abdelhak M. Zoubir · 2018
Distributed signal processing for sensor networks with node-specific interest requires the common labeling of all objects of interest. Current methods formulate the labeling task as a data clustering problem after extracting source-specific features. They assume perfect knowledge of the number of clusters, which is mostly unavailable and possibly time-varying. Thus, we propose distributed and adaptive Bayesian cluster enumeration algorithms by extending our recently proposed single node methods to a distributed sensor network setup where the nodes exchange information via the diffusion principle. The proposed methods are applied to a camera network use-case, where multiple users film a nonstationary scene from different angles. The number of pedestrians is estimated based on streaming-in feature vectors without assuming prior information, such as known positions of the devices, registration of camera views or the availability of a fusion center. Experimental results show the effectiveness of the proposed methods for synthetic and real data.